New AI Model Could Spot Advanced Heart Failure Before It Becomes a Crisis

Advanced heart failure is one of medicine’s most difficult conditions to catch early. Patients can look stable during a routine visit, even while their heart is struggling in ways that only become obvious during specialized exercise testing. That gap matters because the disease is serious, affecting roughly 200,000 Americans, with a one-year survival rate below 50 percent. Yet fewer than 6,000 patients a year receive advanced treatments such as transplants or mechanical pumps.

Now, researchers from Weill Cornell Medicine, Cornell Tech, Columbia University Vagelos College of Physicians and Surgeons, and NewYork-Presbyterian are testing whether artificial intelligence can help close that gap using information already collected in ordinary care. Their goal is to identify patients at high risk much earlier, before the condition reaches its most dangerous stage.

How the model works

The study focuses on echocardiograms, the ultrasound scans that are already common in cardiology. On their own, standard echocardiograms have not been strong enough to predict outcomes reliably. But the new model was designed to do something more ambitious: combine several imaging streams at once, including heart chamber motion, valve movement, and Doppler blood-flow signals, and then merge that with clinical data from electronic health records such as age, body mass index, and other routine measurements.

The model was trained on data from 1,000 patients treated at NewYork-Presbyterian/Columbia University Irving Medical Center and then tested on 127 patients across three hospitals. That external testing matters because it checks whether the system still works outside the place where it was developed. In the reported results, the model reached about 85% accuracy in identifying high-risk patients, with strong technical performance scores on both the training hospitals and the separate validation group.

For clinicians, the appeal is obvious: if a patient already has an echocardiogram, the AI could quietly flag concern in the background and prompt a closer look. Instead of waiting for a specialized cardiopulmonary exercise test that many hospitals do not offer, a doctor could receive an early warning during routine care and refer the patient sooner for advanced evaluation.

Why the finding matters

The researchers are careful not to oversell the results. The study showed weaker performance in patients aged 60 and older, and the authors say that may reflect both limited representation in the training data and the greater complexity of disease in older adults. Accuracy also varied across racial groups and imaging types, with Doppler data appearing especially sensitive to differences between hospitals. Those limits suggest the system still needs refinement before it can be widely trusted.

Even so, the study points to a meaningful shift in how heart failure might be detected. Instead of depending mainly on specialized testing at large medical centers, the next generation of diagnosis may begin with everyday scans already being performed in hospitals across the system. If larger prospective trials confirm the findings, AI-assisted echocardiogram review could help identify high-risk patients earlier, improve referrals, and open the door to treatment before the disease becomes irreversible.